Product market competition, corporate governance and firm performance

Research in Economics (2002) 56, 299–332doi:10.1006/reec.2001.0278Available online at http://www.idealibrary.com on

Product market competition, corporategovernance and firm performance: an empiricalanalysis for Germany

SILKE I. JANUSZEWSKI†, JENS KOKE‡ AND JOACHIM K. WINTER§

†Massachusetts Institute of Technology‡Centre for European Economic Research (ZEW), Mannheim§University of Mannheim

(Received 5 March 2001, accepted 13 November 2001)

Summary

This study examines the impact of product market competi-tion and corporate governance on productivity growth in Ger-man manufacturing. Using a panel of almost 500 firms overthe years 1986 to 1994, we find that firms experience higherproductivity growth when operating in markets with intensecompetition. Similarly, productivity growth is higher for firmsunder control of a strong ultimate owner, but not when theultimate owner is a financial institution (a group that consistedalmost exclusively of German banks and insurance firms in oursample period). Our results also indicate that competition andtight control are complements: The positive effect of compe-tition is enhanced by the presence of a strong ultimate owner.

2002 University of Venice

J.E.L. Classification: D24, D43, G32.Keywords: competition, corporate governance, productivity.

1. Introduction

Productivity growth has been slow in many continental Europeancountries such as Germany and France over the last few decades,especially in comparison with the United States. Improving pro-ductivity is important not only to improve firm value, but also froma public policy perspective. For example, social security systemsin many countries move away from pay-as-you-go financing intopartially or fully funded systems, and therefore the rate of returnon capital will be even more important as a determinant of future

1090–9443/02/030299 C 34 $35.00/0 2002 University of Venice

300 S.I. JANUSZEWSKI et al.

generations’ welfare than it has been in the past, as pointed out byBorsch–Supan and Winter (2001).

Many economists argue that there are two main reasons forpoor firm performance in continental European countries, lackof product market competition and poor corporate governance;see Baily and Gersbach (1995), McKinsey Global Institute (1997),Borsch–Supan (1998), and Allen and Gale (2000). A number of the-oretical papers investigate the effects of competition and corporategovernance on firm performance, but the theoretical predictionsare far from unambiguous. Moreover, empirical evidence is sparse,in particular at the level of individual firms. In this paper, we aimto close this gap with an econometric analysis of firm performancein Germany. We use a unique panel data set with detailed infor-mation on almost 500 manufacturing firms over the 1986 to 1994period, and we apply an econometric approach which alleviatesendogeneity problems that typically plague empirical studies offirm performance.

In order to get a preliminary understanding of productivitydifferences between countries, it is helpful to look at how labour,capital and total factor productivity have evolved over time.Figure A shows productivity comparisons between the UnitedStates and the continental European countries of Germany andFrance for 1970 to 1995; the productivity estimates are taken fromMcKinsey Global Institute (1997). The top panels demonstrate thatthe gap in capital productivity between Germany and the UnitedStates has been much larger than the gap in labour productivity.The bottom left panel suggests a possible reason for this: capitalintensity has been higher in Germany than in the United States,favouring relatively high labour productivity, while at the sametime total factor productivity has been low. Taken together, thesefigures show that labour and capital productivity have been higherin the United States than in Germany and France throughoutthis entire period, leading to substantially higher total factorproductivity in the United States (bottom right panel).

The existing empirical evidence also suggests that poor capitalperformance in Germany, France and other continental Europeancountries results in low rates of return on capital (Mueller andYurtoglu, 2000; Borsch–Supan, 1998). In addition, sector-specificproduct market regulations in many European countries constrainexposure to international competition and cause firms to innovateless and rely on less efficient production processes, which alsoreduces productivity growth (see Baily and Gersbach, 1995).

So far, little empirical work has been done on the interactionof corporate governance and product market competition indetermining productivity at the firm level. Nickell et al. (1997)are the first to analyse this question directly using a panel of U.K.firms. Bottasso and Sembenelli (2001) investigate the relationship

CORPORATE GOVERNANCE AND FIRM PERFORMANCE 301

FIGURE A. Productivity trends in France, Germany, and the United States.

between market power and factor productivity in Italy. To ourknowledge, there is no other formal firm-level study of productivitygrowth for bank-based continental European economies in whichcorporate governance problems are judged to be more severe thanin market-based economies such as the United Kingdom or theUnited States. (Edwards and Fischer, 1994).

In this paper, we follow the approach of Nickell et al. (1997)and analyse the role of product market competition and corporategovernance as determinants of productivity growth in Germanmanufacturing firms. We restrict our attention to the manufac-turing sector, because productivity dynamics in service industriesare much more difficult to measure and to estimate in empiricalmodels. Our empirical approach improves on existing empiricalstudies on corporate governance in Germany, such as Cable (1985)and Edwards and Nibler (2000), in two important respects, a verybroad sample of firms and an econometric technique that controlsfor endogeneity problems.

In contrast with much of the existing empirical literature oncorporate governance, with the exceptions of Nickell et al. (1997)


and Bottasso and Sembenelli (2001), our econometric approachallows us to control for the endogeneity of both product marketcompetition and corporate governance. We estimate productivitygrowth equations using a fixed-effects generalized method ofmoments estimator introduced by Arellano and Bond (1991). Whilesuch an approach certainly does not solve all potential econometricproblems, we believe that it is a step in the right direction.

Our main findings are (i) firms experience higher productivitygrowth when operating in markets with intense competition; (ii)productivity growth is higher for firms under control of a strongultimate owner, but not when the ultimate owner is a financialinstitution; and (iii) competition and tight control are complements,that is, the positive effect of competition is enhanced by thepresence of a strong ultimate owner.

The remainder of this paper is structured as follows. In Section 2,we review the existing theoretical and empirical literature onthe relationship between product market competition, corporategovernance and firm performance. We then describe the firm-leveldata used in our empirical analysis (Section 3). In Section 4, wepresent some preliminary evidence on the relationship of productmarket competition, corporate governance and firm productivity.Section 5 contains our empirical model of productivity growth anddetails of the generalized method of moments (GMM) estimationprocedure. In Section 6, we present our estimation results. Section 7concludes.

2. Literature on the determinants of firm productivity

The aim of corporate governance is to overcome incentive prob-lems generated by the separation of ownership and control innon-owner operated firms. The main mechanisms by which cor-porate governance works are internal control and managementcompensation, designed to align managers’ incentives with theowners’ goals (see Shleifer and Vishny, 1997). However, corporategovernance may not always be effective. Factors like concentrationof share ownership, capital structure and board structure deter-mine the owners’ incentives to incur the cost of installing effectivegovernance mechanisms.

Even in the presence of weak corporate governance, fierce prod-uct market competition may act to align managers’ goals with theaim of efficient production; Allen and Gale (2000) provide a review.For example, Holmstrom (1982) and Nalebuff and Stiglitz (1983)show in a setting with cost functions which are stochastic but iden-tical across agents (or firms, respectively) that perfect competitionreveals full information to the owners about the realization of com-mon cost shocks. In this case, relative performance evaluation can


induce optimal management performance. Hart (1983) comparestwo situations with independent and with positively correlated costfunctions, respectively, showing that, when costs are correlated,competition reduces the amount of managerial slack. However,Scharfstein (1988) shows that Hart’s results depend on the extentof managers’ assumed risk aversion. Scharfstein presents a modelthat is similar to Hart’s but assumes a different utility func-tion. The result is contrary to Hart’s: increased product marketcompetition leads to more slack.

Hermalin (1992) identifies four mechanisms of the influence ofproduct market competition on management performance. Theseare an income effect of reduced profits in a more competitiveenvironment, a risk-adjustment effect if profit risks vary with thedegree of competition, an effect of change in returns to managerialeffort, and the effect of improved information in the presenceof more rival firms. None of these effects is unambiguous inthe sense of either reducing or increasing agency costs withincreased competition, but Hermalin shows that if the incomeeffect is positive (i.e. if agency goods are normal) and dominatesthe other three effects, then agency costs decrease with intensifiedcompetition. Schmidt (1997) analyses the impact of competitionon the threat of bankruptcy and resulting managerial effort.He finds that competition increases the likelihood of bankruptcyfor inefficient firms and should, therefore, improve managementperformance.

In summary, theoretical analyses of the effects of product marketcompetition and corporate governance on firm performance do notprovide us with a clear prediction for the sign of these effects.Turning to the existing empirical literature, Caves (1992), Greenand Mayes (1991), Blundell et al. (1995), and Nickell (1996) presentevidence that increased product market competition is associatedwith higher firm productivity or higher productivity growth. Ina recent study with Norwegian establishment-level data, Klette(1999) provides evidence for a negative relationship between price-cost margins, as a measure for product market competition, scaleeconomies and productivity.

The evidence on the effect of shareholder control on productivityis mixed; see Short (1994) for a survey of the empirical literatureon ownership and firm performance. In a recent empirical studyof the relationship between managerial ownership and firmperformance, Himmelberg et al. (1999) stress that both ownershipand performance are determined endogenously by changes inthe firm’s contracting environment. Controlling for observed firmcharacteristics and firm fixed effects, they find no evidence forthe notion that changes in managerial ownership affect firmperformance.


Finally, there is a small, but growing empirical literatureon the interaction of corporate governance and product marketcompetition in their influence on firm performance. Nickellet al. (1997) estimate the effect of product market competition,shareholder control, and debt levels on firm-level productivitygrowth in the United Kingdom, including interaction terms. Theirresults show a positive influence of product market competition,ownership control and financial pressure on productivity growth.In addition, they find evidence that both competition in productmarkets and financial pressure can substitute for internal control.In a similar framework, Nickell and Nicolitsas (1999) investigatehow financial pressure affects firm behaviour, also using datafrom the United Kingdom. They find that an increase in financialpressure has a large negative effect on employment while it hasa small positive effect on productivity. Bottasso and Sembenelli(2001) exploit the impact of the EU Single Market Program toassess the relationship between market power and productivity.Their main finding is that after an exogenous reduction in marketpower, productivity increases.

To our knowledge, there are no other empirical studies onthe interaction of product market competition and corporategovernance for continental European countries. Many studies (e.g.Edwards and Fischer, 1994) argue that control is much tighterin market-based than in bank-based economies. It is thereforeinteresting to compare the findings obtained by Nickell et al.(1997) for the United Kingdom—a market-based economy—withour empirical findings for Germany, an economy with a bank-based system of corporate governance and a complex ownershipstructure characterized by cross-holdings and poor external controlstructures. Koke (2001) provides descriptive evidence on ownershipstructures in German manufacturing.

3. Data and measurement issues

The sample used in this analysis is based on firm-level data forthe years 1986 to 1994. The unbalanced panel comprises 491German firms that operate in the manufacturing sector, with atotal of 3465 firm years. In contrast to previous empirical studieson corporate governance in Germany (e.g. Cable, 1985; Gortonand Schmid, 2000; Becht and Bohmer, 2001) we do not restrictour analysis to large firms listed on the stock exchange, but alsoinclude non-listed firms. This is significant because it alleviatesthe selection bias caused by restricting the analysis to listed firms.Companies from former Eastern Germany are included only after1990. In general, we follow Nickell et al. (1997) and Nickell andNicolitsas (1999) in the construction of our variables, with some


adjustments to take account of data availability and institutionalfactors in Germany. In this section, we explain the composition ofour sample and the economic principles that guide the constructionof the variables used in our empirical analysis; precise definitionsof these variables can be found in the data appendix.

3.1. DATA SOURCES

The analysis of product market competition and corporate gover-nance and their impact on productivity growth is based on threesources of data. The first main pillar—balance sheet data usedto estimate productivity growth—comes from Hoppenstedt’s Bal-ance Sheet Database (henceforth, BSD). An important featureof this data source is that it contains information on listed andnon-listed corporations, both public (Aktiengesellschaft, AG) andprivate (Gesellschaft mit beschrankter Haftung, GmbH). We take1986 as the starting year because a change in disclosure rulesmakes data from annual reports before and after the year 1986incompatible.† The last year of the sample is 1994 because pub-lication of our main source of ownership data, to be describedshortly, was discontinued in that year. For the period 1986 to1994, BSD contains 5222 firms (22 732 firm years) for which con-solidated balance sheet data are available. We eliminate all firmsthat do not operate primarily in the manufacturing sector becauseproductivity in industries such as financial (bank or insurance) ornon-financial services (wholesale or retail trade) is hard to com-pare with productivity in manufacturing. We also eliminate firmsoperating in the utility, traffic and telecommunications industries,which were predominantly government-owned during the periodof observation. Selection by industry leaves us with data on 1799firms.

The second main pillar—data on ownership structures—isobtained from annual reports published by former BayerischeHypotheken- und Wechsel-Bank (in short, Hypobank). Thesereports contain information on direct ownership of common stockfor all listed and large non-listed German corporations. Hypobankreports the size and the name of a direct owner when the sizeof the ownership block exceeds 5%. However, the Hypobankdata on direct ownership rights cannot readily be used becauseownership complexity of German firms requires to examineultimate firm ownership, as pointed out by Koke (2001). Therefore,

† In 1985 several changes were introduced in German corporate law (§289HGB), most of them triggered by the European Community’s Fourth CompanyLaw Directive on the harmonization of national requirements pertaining tofinancial statements.


this study reconstructs voting rights information in a bottom-up approach from information on direct ownership rights (seeSection 3Ð3).

After matching ownership data, we are left with a sample of 1057firms. Because of missing values for important balance sheet items,another 140 firms must be eliminated. This selection proceduregenerates a sample of 917 firms (4001 firm years) with at least oneyear of balance sheet and ownership data during the years 1986to 1994. Since the dynamic panel estimator, which we apply inthe empirical analysis, requires at least three consecutive years ofdata, we further eliminate 426 firms for which we have less thanthree years of consecutive data. Almost all of these deleted firmshave data only for the year 1994 since Hoppenstedt substantiallyincreased firm coverage in 1994. The final sample contains 491firms (3465 firm years).

The third main pillar of data—measures of product marketcompetition—comes from several sources. Information on firm con-centration at the four-digit industry level is obtained from biennialreports of the Federal Anti-Trust Commission (Monopolkommis-sion, 1996). Information on the value of imports and domesticproduction at the two-digit industry level is obtained from theFederal Statistical Office, from Außenhandelsstatistik and Produk-tionsstatistik (Fachserie 4, Reihe 3Ð1), respectively. On the basisof these industry-level data we construct measures of competition,domestic as well as from abroad. In addition, we construct a firm-specific measure of competition based on balance sheet data (seeSection 3Ð2).

The sample is fairly representative for the universe of largeGerman corporations. Taking the number of all incorporatedGerman firms in the year 1992 as a reference, coverage is highfor listed firms (48Ð9%), all of which are public corporations. Thesample includes all firms listed on any German stock exchangethat mainly operate in the manufacturing sector. For non-listedfirms, coverage is small for public corporations (8Ð9%) and weakfor private corporations (0Ð02%). Controlling for firm size, thesample includes 66Ð1% of all public corporations with total salesexceeding 100 million DM, and more than 3% of large privatecorporations.

Sample attrition is a concern since it might result in selectionbiases. To test for a potential selection bias, we analyse informationon firms’ survival status. For firms leaving the sample before 1994,information is obtained from BSD and telephone interviews. Wefind that 31 out of 54 firms that exit the sample before 1994 stillexisted in 1994 without a change in ultimate ownership—theysimply changed their name or stopped reporting owing to reasonsdetermined within the firm. In 16 cases, operation was shut downdue to liquidation or bankruptcy. In 7 cases, a firm had been taken


over by another entity. Hence, the majority of firm exits during thesample period are not related to firm failure or acquisition.†

3.2. MEASURING COMPETITION

The main variable used to measure competition is the firm’s rentsfrom production (RENT), which can be interpreted as an ex postmeasure of market power. The motivation for using this measureis that firms operating in less competitive markets should be ableto sell their products well above marginal cost, and therefore earnhigher rents after covering their expenses (on capital, labour andmaterials). The abstract definition of production rents, Rt, is asfollows:

Rt D St � rKt pK

t Kt

Qt. .1/

The denominator, Qt, is real output (total sales), pYt Yt.‡ The

numerator is a measure of the firm’s real operating surplus, St, lessreal cost of capital, rK

t pKt Kt. In this notation, Yt is nominal output,

Lt, Kt, and Mt are nominal labour, capital and materials inputs,while pY

t , pLt , pK

t and pMt are the corresponding prices. Finally, rK

tis the user cost of capital.

In the literature (e.g. Nickell, 1996), raw operating surplus, St,is measured by ‘‘earnings before interest, taxes and depreciation’’(also known as EBITDA). This quantity contains a number ofbalance-sheet items that can potentially distort the economiccontent of this variable, resulting in values of EBITDA that aredownward-biased measures of raw operating surplus. This problemis particularly severe in Germany, where firms are entitled toretain a large fraction of earnings to build up reserves. In oursample, this effect is large enough to make the mean of the rentsvariable negative in the pooled sample, with the implication that,loosely speaking, a large number of firms make losses most of thetime. We therefore do not use balance-sheet EBITDA as a measureof raw operating surplus. Instead, we use an economic definitionof raw operating surplus: sales less costs for materials and labour,hence St D pY

t Yt � pMt Mt � pL

t Lt. In economic terms, this definitionis equivalent to the definition of EBITDA.

The user cost of capital is given by rKt D dC rt, where d is the

depreciation rate and rt is the risk-free market interest rate. An

† A more detailed analysis of sample representativeness and of potential biasesintroduced by our sample selection procedure is available on request. In addition,see Koke (2000) for an analysis of selection, entry, and attrition biases in a panelof firms that is identical to the one we use.

‡ We have also used real value added, pYt Yt � pM

t Mt, in the denominator tocheck for robustness. All results reported below remain qualitatively unchanged.


alternative definition, which follows Nickell et al. (1997), alsoincludes a firm-specific risk premium, Qrt, implying rK

t D dC rt C Qrt.In analogy to Nickell (1996), the risk premium is equal to thefirm’s average stock market return over the period 1986 to 1994less the average short-term interest rate over the same period. Fornon-listed firms, we use the average industry- and year-specificrisk premium that is observed for listed firms. In this study, wereport results using both definitions: RENT is rents without riskadjustment, while RENTR uses the risk-adjusted interest rate.With this in mind, the abstract definition of firm rents in (1) canbe rewritten in terms of observable quantities as follows:

Rt D .pYt Yt � pM

t Mt � pLt Lt/� rK

t pKt Kt

Qt. .2/

In addition to firm-specific rents, we use the market share of the sixlargest producers (CR6) and the respective Herfindahl index (HHI),both measured at the four-digit industry level, as proxy variablesfor competition. As a proxy for competition from abroad we usethe ratio of imports to total market size (i.e. the sum of domesticproduction and imports), measured at the two-digit industry level(IMPORT).

There are two important caveats with respect to all measuresof competition used in this paper. Firstly, we acknowledge thatthese variables do not reflect some important facets of competition,namely, potential entry and firm conduct. Secondly, as we do nothave firm-specific data on market shares, we can only assigncompanies to their primary four-digit industry group, but wecannot adjust Herfindahl indices and concentration ratios usingfirms’ market shares.

To illustrate our measures of competition, Table 1 describes theintensity of competition using all of these measures, separately forthe 22 two-digit manufacturing industries contained in the sample.We find that German manufacturing firms earned rents of about15% during the years 1986 to 1994. On average, the six largestproducers cover more than one-third of the domestic market, andimports make up for about one-fourth of the total market. However,Table 1 indicates some remarkable differences between industries.Imports represent a large fraction of the total market in textiles,clothing, leather, machines for data processing and other vehicles(e.g. ships). Correspondingly, the market share of the six largestproducers is low in textiles and machines for data processing, buthigh for tobacco. Industries with low RENT are textiles, metals andother vehicles. As we could expect, RENT is negatively correlatedwith industry concentration and import penetration. However,this correlation is weak. This implies that the empirical analysisshould include firm-level as well as industry-level measures ofcompetition.


TABLE 1 Firm and industry-specific measures of competition

Firm level Industry level Observations

RENT CR6 HHI IMPORT Number Percent

Food (15) 23Ð3% 23Ð2% 1Ð6% 18Ð6% 523 15Ð1%Tobacco (16) 22Ð3% 97Ð6% 20Ð0% n.a. 47 1Ð4%Textiles (17) 7Ð9% 36Ð5% 3Ð6% 45Ð9% 188 5Ð4%Clothing (18) 24Ð7% 12Ð7% 0Ð6% 58Ð2% 67 1Ð9%Leather (19) 18Ð2% 35Ð2% 3Ð5% 60Ð4% 17 0Ð5%Wood (20) 16Ð2% 26Ð5% 2Ð2% 24Ð2% 20 0Ð6%Paper (21) 14Ð0% 20Ð9% 1Ð6% 27Ð8% 113 3Ð3%Publishing, 16Ð1% 15Ð1% 0Ð7% 6Ð4% 26 0Ð8%

printing (22)Coal, oil 13Ð5% 82Ð1% 12Ð9% 41Ð9% 32 0Ð9%

processing (23)Chemicals (24) 19Ð7% 38Ð6% 3Ð5% 25Ð4% 308 8Ð9%Rubber, plastic 14Ð8% 46Ð3% 5Ð3% 19Ð3% 212 6Ð1%

products (25)Rock, stone, 17Ð0% 61Ð4% 9Ð8% 17Ð7% 231 6Ð7%

glass (26)Metals (27) 4Ð5% 35Ð5% 3Ð3% 22Ð7% 195 5Ð6%Metal 14Ð9% 29Ð7% 2Ð5% 14Ð5% 179 5Ð2%

products (28)Machinery (29) 12Ð8% 34Ð7% 3Ð1% 16Ð6% 599 17Ð3%Equ. for data 14Ð1% 8Ð3% 0Ð3% 57Ð1% 132 3Ð8%

processing (30)Equ. for power 13Ð4% 42Ð6% 4Ð7% 27Ð7% 126 3Ð6%

generation (31)Equ. for 11Ð9% 42Ð3% 5Ð3% n.a. 87 2Ð5%

broadcasting andTV (32)Medical and 15Ð6% 37Ð9% 4Ð7% 36Ð2% 69 2Ð0%

opticalinstruments (33)Cars, car 11Ð3% 35Ð5% 3Ð1% 20Ð0% 156 4Ð5%

parts (34)Other vehicles (35) 4Ð6% 55Ð8% 7Ð1% 70Ð7% 86 2Ð5%Furniture, jewelry, 22Ð4% 35Ð6% 3Ð8% 25Ð8% 52 1Ð5%

toys (36)Total 15Ð2% 36Ð0% 3Ð9% 25Ð7% 3465 100Ð0%Correlation withRENT

1Ð000 �0Ð060�0Ð024 �0Ð082 — —

Notes: Firm and industry-specific measures of competition, separately by two-digit industry (NACE codes are given in parentheses): the ratio of total operatingsurplus less costs of capital (without risk premium adjustment) to total sales(RENT), the market share of the six largest producers (CR6), the Herfindahlindex of producer concentration (HHI), and the ratio of imports to total marketsize (IMPORT). The sample comprises 491 firms.


3.3. MEASURING CORPORATE GOVERNANCE

The main variable used to measure corporate governance is anindicator of whether a firm has an ultimately controlling owneror not (CONTROL). This measure takes into account complexownership structures, which are frequently encountered in largeGerman firms. Therefore, it is not only based on measures ofdirect ownership, which can be misleading, in particular, forconglomerates (Koke, 2000; Becht and Bohmer, 2001). Similarly, itclearly identifies one single owner. This allows us to classify firmsaccording to the type of their ultimate owner.

The identification of the ultimate owner for each firm is based onGerman corporate law and involves two steps. Firstly, we identifythe ultimate owner for each direct shareholder using the followingthree rules. Rule 1 (strong ownership rule): A chain of controlis pursued to the next level if the shareholder being analysed isowned to 50% or more by a shareholder on the next level, whileall other shareholders on the next level own less than 50%. Rule 2(weak ownership rule): If rule 1 does not apply, a chain of controlis pursued to the next level if the shareholder being analysed isowned to 25% or more by a shareholder on the next level, whileall other shareholders on the next level own less than 25%. Rule 3(stop rule): If neither rule 1 nor rule 2 applies, a chain of controlis not pursued further. These rules guarantee that no more thanone ultimate owner is identified for each direct shareholder. Notethat if a shareholder has split his ownership stake in a particularcompany into several smaller stakes, for example into two blocks of50% held by two subsidiary firms, we combine these smaller stakesinto one single block. We set the first cut-off point at 50% becauseGerman law allows an investor owning 50% of all shares to appointmanagement.† The second cut-off point is set at 25% because aninvestor owning 25% of the shares has the right to veto decisions.In a second step in determining the ultimate owner for each samplefirm, we apply the three rules to all direct shareholders. This allowsus to identify one single shareholder that is in ultimate control.When no single shareholder fulfills the criteria, this firm is seen tohave no ultimate owner.

To illustrate our measure of corporate governance, Table 2describes how average concentration of ownership evolves overtime. Besides CONTROL, Table 2 also shows two other measures ofownership concentration commonly used in the literature: the sizeof the largest block (BLOCK) and the Herfindahl index (HERF)calculated for all large share blocks. Note that BLOCK as well

† A 50% majority is sufficient to dismiss management after their regular periodof office. But a majority of 75% is required to dismiss management during itsperiod of office (§103 (1) AktG).


TABLE 2 Ownership concentration

Mean Median

CONTROL BLOCK HERF BLOCK HERF

1986 79Ð8% 58Ð9% 46Ð0% 53Ð0% 36Ð0%1987 84Ð1% 65Ð0% 53Ð2% 68Ð0% 50Ð0%1988 85Ð8% 68Ð3% 57Ð5% 75Ð7% 57Ð8%1989 87Ð1% 70Ð8% 60Ð2% 77Ð3% 60Ð3%1990 88Ð8% 71Ð4% 60Ð8% 77Ð3% 60Ð4%1991 89Ð4% 72Ð7% 62Ð2% 79Ð6% 63Ð4%1992 90Ð5% 74Ð7% 64Ð9% 83Ð9% 70Ð3%1993 90Ð5% 75Ð6% 65Ð7% 84Ð0% 70Ð5%1994 89Ð9% 74Ð5% 64Ð5% 83Ð7% 70Ð1%Average 88Ð1% 71Ð4% 60Ð8% 77Ð3% 61Ð1%Correlation with 1Ð000 0Ð639 0Ð499 0Ð639 0Ð499

CONTROL

Notes: Concentration of ownership in a given year for the period 1986 to 1994.Measures of ownership concentration include the average size of the largestshare block (BLOCK), the average Herfindahl index of ownership concentration(HERF), and the fraction of firms for which an ultimate owner can be identified(CONTROL). The sample comprises 491 firms.

as HERF refer to the direct level of ownership. We find thatownership is highly concentrated. Examining ownership at theultimate level, we identify a controlling owner for, on average,88Ð1% of sample firms during the years 1986 to 1994. At thedirect level of ownership, the largest block is also very largewith 71Ð4% at the mean and 77Ð3% at the median. Similarlyhigh concentration is found when using the Herfindahl index.Both BLOCK and HERF are highly correlated with CONTROL.Hence, collinearity problems would be likely when using all threemeasures simultaneously in the empirical analysis. The degreeof complexity in ownership structures makes CONTROL a moreappealing measure of ownership concentration, hence, CONTROLis our preferred measure of corporate governance in this study.

Finally, we include the type of the ultimate owner (TYPE) anda measure of ownership complexity (CROSS). The largest fractionof firms is ultimately controlled by individuals (42Ð2%) and by non-financial firms (36Ð9%). Only 4Ð4% of firms are ultimately controlledby a bank, and 2Ð2% are controlled by other financial institutions.Note that actual voting power of banks might be greater in practicewhen banks make use of proxy voting. However, recent evidencesuggests that proxy voting is extremely unlikely to significantlyenhance bank voting power (Edwards and Nibler, 2000). Only2Ð2% of firms are controlled by government agencies, and 11Ð9% offirms have dispersed ownership. Regarding ownership complexity,7Ð1% of sample firms are controlled by a firm that belongs to the


well-known web of German industrial and financial conglomerates(Wenger and Kaserer, 1998).

4. Preliminary evidence on competition, corporate governanceand productivity

We begin our empirical analysis with some suggestive evidencebased on a simple measure of firm productivity. We estimate astandard two-factor Cobb–Douglas production function with valueadded as the dependent variable (i.e. we indirectly account formaterials as third input factor), and we interpret the residuals fromthis static regression as a measure of relative firm productivity(i.e. relative to the regression mean). The concept of relativeproductivity has a long tradition in applied productivity analysis;see Doms et al. (1995) for an application. Table 3 reports the resultsfrom several specifications. In these regressions, the estimates ofthe unrestricted coefficients do not allow us to reject the hypothesisof constant returns to scale with respect to labour and capital.

To get a first impression of the effects of product marketcompetition and corporate governance on firm productivity, wesplit the sample into two groups: firms with above and firmswith below average productivity. In Table 4, we report means ofsome key measures of competition and corporate governance forboth splits. As a proxy for productivity, we use the residual from

TABLE 3 Cobb–Douglas production function estimates without timedynamics

Independent Dependent variable: Log output .yit/variables Pooled OLS Fixed effects

Log labour .lit/ 0Ð874ŁŁŁ 0Ð951ŁŁŁ 0Ð752ŁŁŁ 0Ð742ŁŁŁ(0Ð015) (0Ð017) (0Ð039) (0Ð039)

Log capital .kit/ 0Ð135ŁŁŁ 0Ð074ŁŁŁ 0Ð337ŁŁŁ 0Ð318ŁŁŁ(0Ð014) (0Ð016) (0Ð063) (0Ð063)

Intercept 9Ð937ŁŁŁ 10Ð507ŁŁŁ 6Ð858ŁŁŁ 7Ð173ŁŁŁ(0Ð202) (0Ð295) (1Ð191) (1Ð189)

Industry dummies no yes no noTime dummies no yes no yesNumber of 3465 3465 3465 3465

observationsR2 0Ð813 0Ð827 0Ð804 0Ð807

Notes: Estimates of a static two-factor Cobb–Douglas production function. Resultsare reported for pooled OLS and fixed effects regressions, with and withoutindustry and time dummies. Asymptotic standard errors (reported in parentheses)are robust to general cross-section and time-series heteroskedasticity.ŁŁŁindicate significance at the 0Ð10, 0Ð05, and 0Ð01 levels, respectively.

CO

RP

OR

AT

EG

OV

ER

NA

NC

EA

ND

FIR

MP

ER

FO

RM

AN

CE

313

TABLE 4 Bivariate analysis of productivity, competition and corporate governance

Forward-looking horizon/relative performance

0 years 2 years 4 yearsNegative Positive Negative Positive Negative Positive

Panel A: Level of total factor productivityIndustry concentration (HHI) 3Ð9% 3Ð8% 3Ð8% 3Ð9% 3Ð8% 3Ð9%Import penetration (IMPORT) 26Ð1%ŁŁŁ 24Ð8% 25Ð6%ŁŁŁ 23Ð9% 24Ð3%ŁŁ 22Ð6%Rent (RENT) 11Ð1%ŁŁŁ 18Ð5% 11Ð1%ŁŁŁ 17Ð6% 11Ð9%ŁŁŁ 17Ð6%Owner concentr. (CONTROL) 84Ð9%ŁŁŁ 90Ð5% 84Ð0%ŁŁŁ 89Ð5% 82Ð2%ŁŁ 87Ð5%Owner .TYPE D private/ 45Ð6%ŁŁŁ 39Ð0% 46Ð6%ŁŁŁ 38Ð1% 46Ð9%ŁŁŁ 38Ð6%Owner .TYPE D financial firm/ 4Ð8%ŁŁŁ 8Ð4% 4Ð6%ŁŁŁ 8Ð4% 4Ð0%ŁŁŁ 9Ð1%Owner .TYPE D non-fin. firm/ 32Ð1%ŁŁŁ 41Ð5% 30Ð7%ŁŁŁ 41Ð2% 28Ð6%ŁŁŁ 38Ð4%Owner .TYPE D government/ 2Ð3% 1Ð5% 2Ð2% 1Ð7% 2Ð8% 1Ð4%Cross ownership (CROSS) 4Ð9%ŁŁŁ 9Ð0% 4Ð7%ŁŁŁ 8Ð9% 3Ð8%ŁŁŁ 9Ð0%Number of observations 1570 1895 1173 1310 732 811

Continued overleaf

314S

.I.JAN

US

ZE

WS

KI

etal.

TABLE 4 Continued

Forward-looking horizon/relative performance

0 years 2 years 4 yearsNegative Positive Negative Positive Negative Positive

Panel B: Growth of total factor productivityIndustry concentration (HHI) 3Ð9% 3Ð8% 4Ð0%ŁŁŁ 3Ð6% 4Ð2%ŁŁŁ 3Ð4%Import penetration (IMPORT) 25Ð9%Ł 25Ð0% 25Ð5%ŁŁŁ 23Ð8% 24Ð1%ŁŁ 22Ð7%Rent (RENT) 16Ð6%ŁŁŁ 13Ð8% 16Ð0%ŁŁŁ 12Ð9% 15Ð4% 14Ð2%Owner concentr. (CONTROL) 88Ð1% 87Ð5% 86Ð2% 87Ð2% 81Ð5%ŁŁŁ 88Ð3%Owner .TYPE D private/ 42Ð7% 41Ð9% 43Ð1% 42Ð1% 43Ð2% 43Ð1%Owner .TYPE D financial firm/ 6Ð6% 6Ð5% 6Ð6% 5Ð6% 6Ð6% 5Ð6%Owner .TYPE D non-fin. firm/ 37Ð0% 37Ð1% 34Ð4% 37Ð6% 29Ð4%ŁŁŁ 37Ð7%Owner .TYPE D government/ 1Ð7% 2Ð1% 2Ð1% 1Ð9% 2Ð3% 2Ð0%Cross ownership (CROSS) 7Ð0% 7Ð0% 6Ð8% 6Ð2% 6Ð4% 5Ð6%Number of observations 1493 1481 1087 905 622 504

Notes: Bivariate analysis of competition and corporate governance and their relation to productivity and productivity growth. Productivityis approximated by the residual from pooled OLS estimation of a Cobb–Douglas production function with time and two-digit industrydummies. Productivity growth is approximated by the first difference in these residuals. Productivity is measured at three forward-looking time horizons: zero years (residual in year t), two years (average of residuals in years t through tC 2), and four years (average ofresiduals in years t through tC 4). All other variables are observed in year t� 1. The test statistics are heteroskedastic t-tests of equalmeans.Ł, ŁŁ, ŁŁŁ indicate significance at the 0Ð10, 0Ð05, and 0Ð01 levels, respectively.


the pooled ordinary least squares (OLS) productivity regressionwith industry and time dummies (column 2 of Table 3). PanelA of Table 4 displays the results for the level of productivity,and panel B displays the corresponding results for productivitygrowth, which is approximated by taking the first difference ofthe productivity residuals. The reason for considering both levelsand first differences is that some of our variables for productmarket competition should be highly endogenous to the level ofproductivity. Since productivity growth is less persistent thanproductivity levels, the endogeneity problem may be less severeif lagged values of competition and corporate governance areused; see also the discussion in Nickell (1996). In the econometricanalysis reported below, we use productivity growth as a dependentvariable, and address the potential endogeneity by using aninstrumental variables approach. Here we simply use all variablesthat are supposed to explain productivity with a one-year lag.Taking into account that competition and corporate governancemight affect productivity or productivity growth in the long-runrather than in the short-run, we report results for three differentforward-looking time horizons: zero years, two years, and fouryears. For example, for a time horizon of two years, we calculateproductivity as the average of productivity in year t, year tC 1, andyear tC 2.

We find strong evidence that firms earning higher rents tendto show significantly higher levels of productivity (panel A ofTable 4). Hence, firms facing intense competition appear to be lessproductive. This result holds irrespective of whether we considershort-run or long-run productivity.† We also find that firms areless productive when operating in markets that are subject tointense competition from abroad via imports. Regarding the type ofultimate owner, we find that a significantly larger fraction of firmsunder control of a large private owner belongs to the group of lessproductive firms than to high productive firms. This suggests thatfirms under control of a private owner tend to be less productive.Vice versa, firms under control of a financial or non-financial firmare significantly more productive. Finally, there is strong evidencethat firms being controlled by the well-known web of cross-heldGerman firms (Wenger and Kaserer, 1998) are more productive.

Regarding productivity growth, we again find a strong relation-ship between RENT and our measure of productivity (panel Bof Table 4). However, the suggested impact of RENT is now theopposite: firms earning higher rents, that is, operating in lesscompetitive markets, tend to show significantly lower productivity

† This result also holds when RENT is corrected by a risk premium, assuggested by Nickell et al. (1997), when RENT is measured as the average overtwo consecutive years, or when RENT is measured relative to value added insteadof total sales.


growth. This result is confirmed by an industry-level measure ofcompetition: Firms operating in markets with lower producer con-centration (HHI) tend to see higher productivity growth. This resultholds when industry-level competition is measured by the fractionof the six largest producers (CR6), but not when measured by theimport ratio. Tight control structures (CONTROL) have a positiveeffect on productivity growth only at the long (four-year) horizon.

In summary, the preliminary evidence suggests that in Germanmanufacturing, product market competition and corporate gover-nance are positively related to differences in productivity growth.The observed negative relation between competition and the levelof productivity could, as argued above, result from endogeneity ofrents with respect to the level of productivity, which we believe tobe significant. Note that this purely descriptive analysis ignores allpotential endogeneity problems. In the remainder of this paper, weaddress these problems in a dynamic model of productivity growthestimated with instrumental variable techniques.

5. An empirical model of productivity growth

In this section, we present an empirical model of productivitygrowth derived from the firm’s production function. This modelis augmented with a set of variables that capture the influenceof product market competition and corporate governance onproductivity growth. This approach follows Nickell (1996) andNickell et al. (1997). There are two reasons for estimating growthequations. Firstly, measurement problems are much more severefor levels equations than for growth equations. Secondly, measuresof industry competition might well be endogenous with respect tothe productivity level, but we would argue that this endogeneityproblem is less severe if productivity growth is the dependentvariable.

The starting point of our analysis is a Cobb–Douglas productionfunction with two factor inputs,

Yit D LitbLKit

bK Ait, .3/

where Yit is value added, Lit is labour, Kit is capital, and Ait is ameasure of total factor productivity for firm i in year t. Since weuse value added as the output measure, we implicitly allow formaterials as a third input. We interpret the level of total factorproductivity, Ait, as the compound effect of past variables thatshape a firm’s productivity. After differencing, we can parameterizethese determinants of productivity by level measures of productmarket competition and corporate governance.

In order to arrive at that growth version, we transform theproduction function (3) in several steps. Firstly, we take logs


and include lagged output as an explanatory variable to allowfor endogenous persistence, using a weight l. We also allow forunobserved firm heterogeneity, ai, and include an error term,eit, which is assumed to be serially uncorrelated over time. Thisyields our basic log-linear empirical production function, with smallletters denoting logs:

yit D lyit�1 C .1� l/bLlit C .1� l/bKkit C .1� l/ait C ai C eit. .4/

Secondly, taking first differences eliminates the fixed firm effectai. We obtain the differenced growth version of the Cobb–Douglasproduction function in equation (3):

yit D lyit�1 C .1� l/bLlit C .1� l/bKkit

C .1� l/ait Ceit. .5/

Finally, we parameterize productivity growth (i.e. the firstdifference of total factor productivity) as a linear function of timeeffects .t/, a contemporaneous industry-specific proxy variablethat captures business cycle effects on competition (CYCLE),the lagged firm-specific measure of competition (RENT), a setof lagged industry-specific variables that capture differences inproduct market competition (HHI, IMPORT),† and a set of laggedcorporate governance variables (CROSS, CONTROL, TYPE). Insome specifications, we also include interactions of RENT andthe corporate governance variables. Thus, productivity growth ismodelled as

ait D .tt � tt�1/C b1CYCLEit C b2RENTit�1

C b3.HHI, IMPORT/it�1

C b4.CROSS,CONTROL,TYPE/it�1

C b5[RENTit�1 Ð .CROSS,CONTROL,TYPE/it�1]. .6/

The empirical model of productivity growth is given by equation(5) together with equation (6). The structure of this modelcorresponds to the differenced panel model with lagged endogenousvariables considered by Arellano and Bond (1991). They proposea GMM estimator that allows to exploit lags of the laggeddependent variable as well as lags of the explanatory variablesas instruments. In our application, using this approach addressesthe potential endogeneity problems with respect to the competition

† Of the three industry-level competition variables that we discussed above,we exclude CR6 because of multicollinearity problems. However, all results arequalitatively robust to using CR6 instead of HHI.


and corporate governance variables that enter the right-hand sideof equation (5).†

Arellano and Bond (1991) show that endogenous variables laggedtwo or more periods are valid instruments, provided there is noserial correlation in the time-varying component of the error termsin equation (4); we test this condition for all specifications. Theinstruments we use are yit�j for j ½ 2, lit�2,kit�2, and the secondlags of all time-varying measures of industry-level competitionand of ownership structure. We test for instrument validity usinga Sargan test of over-identifying restrictions.‡ We report thosetests together with the estimation results and standard errors thatare robust with respect to general heteroskedasticity in the nextsection.

While the Arellano-Bond approach can in principle deal withpotential endogeneity problems in our application, there is acaveat. Blundell and Bond (1998) show that in autoregressivemodels with persistent series, the first-difference estimator canbe subject to finite sample bias as a result of weak instruments,and that these biases could be greatly reduced by the inclusionof levels equations in a GMM system estimation procedure (see,e.g. Bond et al., 1999). We do not estimate a model with equationsin both levels and first differences, because of the structure ofour empirical model—as noted before, we assume that the levelof competition influences productivity growth, following Nickell(1996) and Nickell et al. (1997). Hence, while our formulation hasthe advantage that we do not have to compare levels of productivityacross firms and industries, but only changes in productivity, ithas the disadvantage of potentially weak instruments (which is,however, difficult to assess in a given empirical application).

6. Estimation results for the productivity growth models

In this section, we present estimation results for our productivitygrowth model. In Table 5, we report GMM estimates of productivitygrowth equations with various measures of product marketcompetition. These are our industry-level competition variables

† An alternative estimation approach for dynamic panel data models is thestandard IV estimator proposed by Anderson and Hsiao (1981). However, sincewe have modelled the influence of competition and corporate governance onproductivity growth using the parameterization in equation (6), the Anderson-Hsiao IV estimator is not readily applicable in our setting.

‡ Following Arellano and Bond (1991), we use the two-step version of the GMMestimator to obtain the Sargan test statistic, while coefficient estimates are basedon the one-step version. Arellano and Bond report that the one-step Sargan testis sensitive to heteroskedasticity, tending to over-reject the null.

CO

RP

OR

AT

EG

OV

ER

NA

NC

EA

ND

FIR

MP

ER

FO

RM

AN

CE

319

TABLE 5 Effects of competition on productivity growth, GMM results

Independent variables Dependent variable: output growth .yt/

Model (1) Model (2) Model (3) Model (4)

Lagged output growth .yt�1/ �0Ð014 �0Ð002 �0Ð043 0Ð119(0Ð058) (0Ð059) (0Ð064) (0Ð165)

Labour growth .lt/ 1Ð080ŁŁ 1Ð066ŁŁ 1Ð056ŁŁ 0Ð767ŁŁŁ(0Ð474) (0Ð458) (0Ð457) (0Ð147)

Capital growth .kt/ 0Ð668 0Ð724 0Ð627 �0Ð079(0Ð474) (0Ð491) (0Ð459) (0Ð223)

Business cycle .CYCLEt/ 0Ð030 0Ð031 0Ð029 0Ð030(0Ð025) (0Ð026) (0Ð025) (0Ð026)

Industry concentration .HHIt�1/ �11Ð721 �11Ð291 �11Ð587 �12Ð428(14Ð646) (14Ð701) (14Ð540) (13Ð512)

Import penetration .IMPORTt�1/ 1Ð236 1Ð013 1Ð178 0Ð978(1Ð608) (1Ð600) (1Ð592) (1Ð203)

Rent .RENTt�1/ �1Ð449ŁŁŁ(0Ð429)

Continued overleaf

320S

.I.JAN

US

ZE

WS

KI

etal.

TABLE 5 Continued


Model (1) Model (2) Model (3) Model (4)

Rent, risk-adjusted .RENTRt�1/ �1Ð353ŁŁŁ(0Ð404)

Rent, firm average .RENTAFt�1/ �1Ð344ŁŁŁ(0Ð395)

Rent, ind, average .RENTAIt�1/ �3Ð314Ł(1Ð850)

Intercept �2Ð840 �2Ð927 �2Ð532 �2Ð213(2Ð676) (2Ð703) (2Ð338) (2Ð205)

Number of observations 2389 2389 2389 2389Instrument validity (Sargan) P D 0Ð254 P D 0Ð145 P D 0Ð182 P D 0Ð578Second-order correlation of residuals P D 0Ð346 P D 0Ð355 P D 0Ð384 P D 0Ð365Constant returns to scale P D 0Ð271 P D 0Ð258 P D 0Ð263 P D 0Ð249

Notes: GMM regression results using the Arellano and Bond (1991) method. All regressions include time and two-digit industry dummies.Instruments are yit�j for j ½ 2, lit�2, kit�2, and the second lags of all industry-level measures of competition. Asymptotic standard errors(reported in parentheses) are robust to general cross-section and time-series heteroskedasticity.Ł, ŁŁ, ŁŁŁindicate significance at the 0Ð10, 0Ð05, and 0Ð01 levels, respectively.


and different versions of rents, our ex post measure of competitionat the firm level. All competition variables enter with a one-year lag. In model (1), the baseline version, we use our preferreddefinition, RENT, as discussed in Section 3Ð2. Its coefficient issignificantly negative. Recall that rents is an inverse measure ofcompetition—the more intense competition on product markets,the smaller the rents that firms can appropriate. A negativecoefficient therefore implies that more intense competition isassociated with higher rates of productivity growth.

In models (2) through (4), we present results for three alternativedefinitions of the rents variable to check for robustness. In model(2), we use the risk-adjusted measure of firm rents (RENTR), asdescribed in Section 3Ð2. In models (3) and (4), we use two versionsof the rents variable that address potential endogeneity issues.RENTAF is the average of the firm’s rents over the past twoyears. This time aggregation should smooth short-run firm-specificshocks that affect output directly and hence affect rents. RENTAI isthe year-specific average of our firm-specific rents measure acrossthe respective two-digit industry. This cross-sectional aggregationalso wipes out firm-specific shocks, and therefore avoids potentialendogeneity problems associated with the rents variable. We find asignificantly negative effect of rents on productivity growth for allthree alternative definitions. This robustness check gives us someconfidence that our general approach—controlling for endogeneityproblems using an instrumental variables GMM method—isappropriate, and we continue to build our specifications on model(1) in our subsequent analysis.

In terms of substance, our results do not reject the hypothesisthat competition has a positive effect on productivity growth,as judged by our ex post measure of competition. Industry-level measures of competition are, however, not significantlyassociated with productivity growth. The business cycle proxyis also insignificant. These insignificant coefficients might be dueto the fact that time and industry dummies absorb most of thevariation in these industry-level variables. Also, as reported inthe data appendix, we cannot assign industry-level competitionvariables to firms perfectly, because we only have industry codesfor the firms’ primary products.

Note that all versions of our GMM model are generally supportedby the standard battery of specification tests. The Sargan tests donot reject the hypothesis of instrument validity. Also, the tests forsecond-order serial correlation of the residuals do not reject thenull of zero correlation. Wald tests cannot reject the hypothesis ofconstant returns to scale. Finally, in all specifications we report,the slope coefficients and the sets of time and industry dummyvariables are jointly significant according to our Wald tests (notreported). In general, our results for the impact of competition (as


measured by the lagged rents variable) are also robust againstvariations of the lag length chosen for the instruments.

We now turn to the effects of corporate governance on pro-ductivity growth. Table 6 reports regressions that augment thespecification in model 1 of Table 5 with our proxy variables forcorporate governance structures. Note that coefficients of the othervariables do not change qualitatively as a result of adding cor-porate governance variables. In model (5), we begin by addingdummies for the existence of cross ownership (CROSS), and fora highly concentrated ownership structure (CONTROL). The coef-ficients for these variables are not significant. However, once weadd an interaction with rents in model (6), the coefficients for boththe ownership concentration dummy and the interaction term aresignificant, while the coefficient of the rents variable itself becomesinsignificant. These results suggest that firms which have a singleultimate owner, that is, which operate under strong ownership,experience higher productivity growth. This effect is increased bystronger product market competition (i.e. lower values of the rentsvariable).

Since these results indicate that ownership structure appears tobe quite important for productivity growth, we end our analysisby looking more closely at different types of ultimate owners.†We add dummies for the following types of the ultimate owner:private individuals, financial firms and government authorities.The reference category is non-financial firms, while the lastcategory, dispersed ownership, is already accounted for by azero value of the ownership dummy. The coefficients of thedummy variable for the existence of an ultimate owner andits interaction with rents remain significant, and increase inmagnitude. Hence, firms under ultimate control of a non-financialfirm have significantly higher productivity growth than firms withdispersed ownership. Firms under ultimate control of a privateowner or a government agency do not experience significantlydifferent productivity growth relative to the reference group. Incontrast, firms whose ultimate owner is a financial firm havesignificantly lower productivity growth. The interactions with thecompetition measure suggest that productivity growth is higherfor firms owned by the government if there is more competition,but competition influences productivity growth negatively for firmsowned by financial institutions.

These results support the notion that firms with a dominantowner tend to experience higher productivity growth, in particularwhen competition on product markets is strong. However, the oppo-site is true if the ultimate owner is a financial institution—which

† This was suggested by one of the referees.

CO

RP

OR

AT

EG

OV

ER

NA

NC

EA

ND

FIR

MP

ER

FO

RM

AN

CE

323

TABLE 6 Effects of competition and corporate governance on productivity growth, GMM results


Model (5) Model (6) Model (7)Interactions Interactions

RENTt�1 RENTt�1

Lagged output growth .yt�1/ �0Ð019 �0Ð022 �0Ð032(0Ð059) (0Ð058) (0Ð051)

Labour growth .lt/ 0Ð925ŁŁ 0Ð882ŁŁ 0Ð893ŁŁ(0Ð413) (0Ð404) (0Ð441)

Capital growth .kt/ 0Ð553 0Ð483 0Ð641(0Ð481) (0Ð471) (0Ð503)

Business cycle .CYCLEt/ 0Ð031 0Ð031 0Ð032(0Ð025) (0Ð025) (0Ð026)

Industry concentration .HHIt�1/ �11Ð277 �11Ð527 �11Ð416(15Ð092) (15Ð090) (14Ð430)

Import ratio .IMPORTt�1/ 1Ð168 1Ð057 0Ð850(1Ð610) (1Ð613) (1Ð635)

Rent .RENTt�1/ �1Ð400ŁŁŁ �0Ð388 �0Ð394(0Ð426) (0Ð289) (0Ð279)

Cross ownership .CROSSt�1/ �0Ð152 �0Ð152 �0Ð055(0Ð130) (0Ð131) (0Ð075)

Continued overleaf

324S

.I.JAN

US

ZE

WS

KI

etal.

TABLE 6 Continued


Model (5) Model (6) Model (7)Interactions Interactions

RENTt�1 RENTt�1

Owner concentr. .CONTROLt�1/ 0Ð010 0Ð186ŁŁŁ �1Ð069ŁŁŁ 0Ð204ŁŁ �1Ð131ŁŁ(0Ð026) (0Ð058) (0Ð316) (0Ð099) (0Ð512)

Owner .TYPEt�1 D private/ 0Ð006 �0Ð159(0Ð118) (0Ð645)

Owner .TYPEt�1 D financial/ �0Ð454ŁŁ 1Ð832ŁŁ(0Ð234) (0Ð943)

Owner .TYPEt�1 D government/ 0Ð077 �1Ð767Ł(0Ð108) (0Ð973)

Intercept �2Ð672 �2Ð782 �3Ð013(2Ð342) (2Ð359) (2Ð765)

Number of observations 2389 2389 2389Instrument validity (Sargan) P D 0Ð216 P D 0Ð237 P D 0Ð160Second-order correlation of residuals P D 0Ð343 P D 0Ð336 P D 0Ð349Constant returns to scale P D 0Ð442 P D 0Ð542 P D 0Ð406

Notes: GMM regression results using the Arellano and Bond (1991) method. All regressions include time and two-digit industrydummies. Instruments are yit�j for j ½ 2, lit�2, kit�2, the second lags of all industry-level measures of competition, and the second lagsof all time-varying measures of ownership structure. The reference category of owner type is ‘non-financial firms’. Asymptotic standarderrors (reported in parentheses) are robust to general cross-section and time-series heteroskedasticity.Ł, ŁŁ, ŁŁŁ indicate significance at the 0Ð10, 0Ð05, and 0Ð01 levels, respectively.


is not a very favourable account of the bank-based system of cor-porate governance in Germany. We also find significant effects ofproduct market competition on productivity growth. Even thoughour research design is more complex due to differences in insti-tutional arrangements, some of our findings stand in contrast tothose obtained by Nickell et al. (1997) for the market-based U.K.system. Specifically, they find that strong ownership and prod-uct market competition are substitutes in improving productivity,while we find that they are complements. Taken together, thesefindings add weight to existing evidence on the differential perfor-mance of Anglo-Saxon and continental European firms obtained inaggregate or industry-level studies. In general, the results of ourdynamic econometric approach are in line with the preliminaryanalysis in Section 4.

We end this section with a discussion of some robustness checkswe performed. With respect to instrument choice, results donot change qualitatively when we do not use the industry-levelcompetition and ownership variables as instruments or when weuse lags longer than two years for the factor inputs. As in Table 5,using alternative measures of rent does not affect the results inthe specifications reported in Table 6. Using the Herfindahl indexof ownership concentration (HERF) or the size of the largest block(BLOCK) as alternative measures of ownership concentration doesnot produce any significant results. Finally, all of our resultsare robust against alternative definitions of the capital stock. Weexperimented with capital stock measures constructed using themethod of Wadhwani and Wall (1986), and later used by Nickell(1996) and Nickell et al. (1997). They also apply a perpetualinventory method, but they do not assume a constant rate ofdepreciation. We also experimented with annual depreciation ratesof 4% and 12%, but our estimation results turned out to be robust.

7. Conclusion

In this paper, we present an empirical analysis of the effectsof product market competition and ownership control on totalfactor productivity growth using a panel data set of almost 500German manufacturing firms over the 1986 to 1994 period. Wefind a significantly negative impact of rents appropriated by firms,indicating that product market competition has a positive impacton productivity growth. The effects of industry concentrationmeasures are insignificant. In our empirical model, the influenceof competition is measured by (lagged) economic rents, a firm-specific ex post measure. Because of the fact that even laggedeconomic rents are endogenous, we use an instrumental variablestechnique to control for potential endogeneity. The negative impact


of lagged rents on productivity growth turned out to be extremelyrobust against various alternative specifications. We are thereforeconfident that even though the GMM method we apply might besubject to small sample biases, our main findings are reliable.

With respect to ownership structure, we find strong evidence thatfirms under tight control experience higher rates of productivitygrowth. However, when we look at different types of ultimateowners, we find that this result does not hold if the firm isowned by a financial institution. Since we do not find an impactof ownership concentration at the direct level (as opposed toultimate ownership), our results suggest that for productivitygrowth, ultimate ownership matters rather than direct ownership.In empirical studies of corporate governance, measures of directownership—which are of course easier to obtain—might thereforebe misleading. This is in line with findings by Koke (2000).Furthermore, interaction of the variables for ultimate ownershipwith our measure for firm-level competition reveals that thedisciplining effect of product-market competition is enhanced bytighter control structures. Interestingly, firms whose ultimateowner is a financial institution show lower rates of productivitygrowth, even if they operate in markets with strong competition.This result is in contrast to findings by Nickell et al. (1997), butconfirms worries about the role of banks for corporate governancein Germany. More research on this issue is certainly required,in particular, since in recent years financial institutions otherthan banks are becoming more important. Recall that our dataend in 1994, and up to that point, the financial institutionsrecorded as ultimate owners were almost exclusively banksand insurance firms. As pointed out by Borsch–Supan andWinter (2001), the role of investment and pension funds iscurrently increasing, and they might have stronger incentivesto improve corporate governance and to enhance productivitygrowth.

In addition to this characterization of banks’ role in corporategovernance, our results have other interesting policy implications.Our results suggest that while the role of financial institutions asultimate owners is not very favourable, concentrated ownershipthat is often associated with a bank-based system has in itselfa positive impact on productivity growth. Finally, our resultsadd empirical support to some well-known hypotheses about thewelfare effects of product market competition, as discussed byAllen and Gale (2000) and Borsch–Supan (1998). The implicationsfor competition policy within continental Europe and the EuropeanUnion at large are obvious—increased competition in the commonmarket should have positive effects on productivity growth.

We end this paper by discussing a few unresolved issues. First ofall, it might come as a disappointment that other than rents, our ex


post measure of competition, all variables that measure industry-level competition are generally not significant in the productivitygrowth regressions. This might be due to measurement problems.As noted before, the industry classification of firms at the four-digit level according to their primary product is quite unreliable.Also, concentration measures can only indicate potential, notactual competition, whereas ex post rents reflect the effects of themarket power that a firm can actually exert. We would also arguethat not much progress can be made by considering alternativemeasures of the level of competition in a given industry. It seemsmore promising to look at changes in competition triggered byexogenous events, but this approach is more suitable for industry-level studies, than for a sector-wide panel study of manufacturingfirms. While there are, to our knowledge, as yet no such casestudies available for Germany, recent product market deregulationin the German telecommunications and energy markets, and inthe European Union at large, provide data suitable for empiricalstudies in the future, as in Bottasso and Sembenelli (2001).

We conclude this paper by discussing directions for futureempirical research that focus on the role of ownership structure andcorporate governance in Germany. As pointed out by Himmelberget al. (1999), changes in the ownership structure should haveimportant effects on productivity growth. Although these changesare endogenous as well, it should be worthwhile to identifychanges in ownership structure in a panel of firms and investigatetheir effects on firm performance. Recent descriptive evidencefor Germany reported by Koke (2000) suggests that followingmajor changes in ownership, industry-adjusted productivity ofpoorly performing firms improves, but remains below the industrybenchmark. In addition, ownership structure does not only affectproductivity growth, but also firm survival and market exitdecisions. To our knowledge, there are no empirical studies thatinvestigate any of these issues in a systematic fashion for Germany,and they are well worth being explored.

Finally, there might be much more complex interactions ofownership structure, board structure, capital structure, and firmperformance. Following Jensen (1986), a high degree of leverageis regarded as a device of disciplining management. Chevalier(1995) finds that highly leveraged firms are weak competitorsin the product market. Similarly, the evidence presented inKovenock and Phillips (1997) and Zingales (1998) indicates thatthe interaction of firm leverage and product market competitionis important in determining future firm performance. All theseempirical studies use data from the United States. For the UnitedKingdom, results by Nickell et al. (1997) and Nickell and Nicolitsas(1999) suggest that financial pressure has a positive effect onproductivity growth. It would be interesting to investigate whether


a similar mechanism is at work in Germany’s bank-based system ofcorporate governance. For example, Borsch–Supan (1998) reportsthat in Germany, creditors do not use their control potentialto improve firm productivity. In future empirical work, directmeasures of bank influence should therefore be used to investigatethe role of banks, although such measures are hard to obtain.

Acknowledgements

We wish to thank Fred Ramb for providing some of the data usedin this paper and Dominik Krupp for valuable research assistance.Axel Borsch–Supan, Herbert Buscher, Dietmar Harhoff, NorbertJanz, Melanie Luhrmann, Anja Wodrich, three anonymous ref-erees, the associate editor and seminar participants at the 2000annual meetings of the European Economic Association and ofthe Verein fur Socialpolitik provided helpful comments on earlierdrafts of this paper. All remaining shortcomings are ours. Financialsupport by Deutsche Forschungsgemeinschaft through grant No.BO 934/7-1 and Sonderforschungsbereich 504 at the University ofMannheim is gratefully acknowledged.

References

Allen, F. & Gale, D. (2000). Corporate governance and competition. In X. Vives,Ed., Corporate Governance: Theoretical and Empirical Perspectives. Cambridge,U.K.: Cambridge University Press.

Anderson, T.W. & Hsiao, C. (1981). Estimation of dynamic models with errorcomponents. Journal of the American Statistical Association, 76, 598–606.

Arellano, M. & Bond, S. (1991). Some tests of specification for panel data: MonteCarlo evidence and an application to employment equations. Review of EconomicStudies, 58, 277–297.

Baily, M.N. & Gersbach, H. (1995). Efficiency in manufacturing and the need forglobal competition. Brookings Papers on Economic Activity—Microeconomics,307–347.

Becht, M. & Bolnner, E. (2001). Voting control in German corporations.International Review of Law and Economics, forthcoming.

Blundell, R. & Bond, S. (1998). Initial conditions and moment restrictions indynamic panel data models. Journal of Econometrics, 87, 115–143.

Blundell, R., Griffith, R. & Reenen, J.V. (1995). A dynamic count data model oftechnological innovation. Economic Journal, 105, 333–344.

Bond, S., Harhoff, D. & van Reenen, J. (1999). Investment, R&D and financialconstraints in Britain and Germany. (unpubl.) manuscript, Institute for FiscalStudies, London.

Borsch–Supan, A.H. (1998). Capital productivity and the nature of competition.Brookings Papers on Economic Activity—Microeconomics, 205–244.

Borsch–Supan, A.H. & Winter, J.K. (2001). Population aging, savings behaviorand capital markets. Working Paper No. 8561, National Bureau of EconomicResearch (NBER), Cambridge, MA.

Bottasso, A. & Sembenelli, A. (2001). Market power, productivity and the EUsingle market program: evidence from a panel of Italian firms. EuropeanEconomic Review, 45, 167–186.


Cable, J. (1985). Capital market information and industrial performance: the roleof West German banks. Economic Journal, 95, 118–132.

Caves, R.E. (1992). Industrial Efficiency in Six Nations. Cambridge, MA: MITPress.

Chevalier, J.A. (1995). Capital structure and product-market competition:empirical evidence from the supermarket industry. American Economic Review,85, 415–435.

Doms, M., Dunne, T. & Roberts, M.J. (1995). The role of technology use inthe survival and growth of manufacturing plants. International Journal ofIndustrial Organization, 13, 523–542.

Edwards, J. & Fischer, K. (1994). Banks, Finance and Investment in Germany.Cambridge, U.K.: Cambridge University Press.

Edwards, J. & Nibler, M. (2000). Corporate governance in Germany: the role ofbanks and ownership concentration. Economic Policy, 15, 237–267.

Gorton, G. & Schmid, F.A. (2000). Universal banking and the performance ofGerman firms. Journal of Financial Economics, 58, 29–80.

Green, A. & Mayes, D. (1991). Technical inefficiency in manufacturing industries.Economic Journal, 101, 523–538.

Hart, O.D. (1983). The market mechanism as an incentive scheme. Bell Journalof Economics, 14, 366–382.

Hermalin, B.E. (1992). The effects of competition on executive behavior. RANDJournal of Economics, 23, 350–365.

Himmelberg, C.P., Hubbard, R.G. & Palia, D. (1999). Understanding thedeterminants of managerial ownership and the link between ownership andperformance. Journal of Financial Economics, 53, 353–384.

Holmstrom, B. (1982). Moral hazard in teams. Bell Journal of Economics, 13,324–340.

Jensen, M.C. (1986). Agency costs of free cash flow, corporate finance, andtakeovers. American Economic Review, Papers & Proceedings, 76, 323–329.

Klette, T.J. (1999). Market power, scale economies and productivity: estimatesfrom a panel of establishment data. Journal of Industrial Economics, 47,451–476.

Koke, J. (2000). Control transfers in corporate Germany: their frequency, causesand consequences. Discussion Paper No. 00-67. Zentrum fur EuropaischeWirtschaftsforschung (ZEW), Mannheim.

Koke, J. (2001). New evidence on ownership structures in Germany. Kredit undKapital, 34, 257–292.

Kovenock, D. & Phillips, G.M. (1997). Capital structure and product marketbehavior: an examination of plant exit and investment decisions. Review ofFinancial Studies, 10, 767–803.

McKinsey Global Institute (1997). Removing Barriers to Growth and Employmentin France and Germany. Frankfurt, Paris, Washington: McKinsey GlobalInstitute.

Monopolkommission (1996). Wettbewerbspolitik in Zeiten des Umbruchs:Hauptgutachten 1994/1995. Baden-Baden: Nomos.

Mueller, D.C. & Yurtoglu, B.B. (2000). Country legal environments and corporateinvestment performance. German Economic Review, 1, 187–200.

Nalebuff, B.J. & Stiglitz, J.E. (1983). Information, competition, and markets.American Economic Review, Papers & Proceedings, 73, 278–283.

Nickell, S. (1996). Competition and corporate performance. Journal of PoliticalEconomy, 104, 724–746.

Nickell, S. & Nicolitsas, D. (1999). How does financial pressure affect firms?European Economic Review, 43, 1435–1456.

Nickell, S.J., Nicolitsas, D. & Dryden, N. (1997). What makes firms perform well?European Economic Review, 41, 783–796.

Scharfstein, D. (1988). Product-market competition and managerial slack. RANDJournal of Economics, 19, 147–155.


Schmidt, K. (1997). Managerial incentives and product market competition.Review of Economic Studies, 64, 191–213.

Shleifer, A. & Vishny, R.W. (1997). A survey of corporate governance. Journal ofFinance, 52, 737–783.

Short, H. (1994). Ownership, control, financial structure and the performance offirms. Journal of Economic Surveys, 8, 203–249.

Wadhwani, S. & Wall, M. (1986). The U.K. capital stock: new estimates ofpremature scrapping. Oxford Review of Economic Policy, 2, 44–55.

Wenger, E. & Kaserer, C. (1998). The German system of corporate governance: amodel which should not be imitated. In S. Black & M. Moersch, eds, Competitionand Convergence in Financial Markets: The German and Anglo-AmericanModels. Amsterdam: Elsevier.

Zingales, L. (1998). Survival of the fittest or the fattest? Exit and financingdecisions in the trucking industry. Journal of Finance, 53, 905–938.

Appendix: Definition of variables

The variables we use in our econometric analysis are constructedin analogy to Nickell et al. (1997) wherever possible, in order toallow direct comparison with their results. Minor differences occurbecause we choose a slightly different method to construct ourcapital stock variable (see below). All variables used in this studyare appropriately deflated and measured in prices of 1991. Sourcesof price and cost indexes and other aggregate variables are givenbelow, together with details on how we constructed each variableused in the empirical analysis.

VALUE ADDED

The firm’s value added, Yt, is defined as output (total sales) lesstotal materials costs. Real values are obtained using a two-digitindustry-specific producer price index published by the FederalStatistical Office (Statistisches Bundesamt, Fachserie 17, Reihe 2,1998) for output, and a combined input price index for materials.The latter does not vary by industry.

CAPITAL STOCK

The firm’s capital stock, Kt, is defined as replacement costs oftangible assets including machines, buildings and land, deflatedusing a combined input price index for capital goods andland, weighted by their empirical distribution (StatistischesBundesamt, Fachserie 17, Reihe 2, and Fachserie 17, Reihe 4,1998). Replacement costs of capital are calculated using the methodof Bond et al. (1999). They adjust the historical cost values forinflation and then apply a perpetual inventory method with a


constant annual depreciation rate of d D 0Ð08. Specifically.

pKt Kt D .1� d/pK

t�1Kt�1pK

t

pKt�1C pK

t It,

where Kt is the capital stock, pKt is the price index for capital goods,

It is real investment and d the depreciation rate. The starting valueis the net book value of tangible assets, adjusted for inflation inprevious years.

LABOUR

The firm’s labour input, Lt, is defined as the total number ofemployees.

BUSINESS CYCLE PROXY

To control for business cycle effects, we use a survey-based indexof capacity utilization at the two-digit industry level as a proxyvariable (CYCLE). This index is part of the ifo Geschaftsklimaand was obtained from the ifo Institut fur Wirtschaftsforschung,Munich.

INDUSTRY-LEVEL COMPETITION

We use three measures to proxy for industry-level competition.As a measure of foreign competition, we use import penetration(IMPORT), defined as the ratio of the total value of importsto total market size. The latter is the sum of imports anddomestic production, measured at the two-digit industry-level.Regarding industry concentration, we use the market shareof the largest six producers, CR6, and the Herfindahl indexof producer concentration. HHI, both of which are measuredfor four-digit output classes. This information is obtained frombiennial reports of the German Federal Antitrust Commission,as reported in Monopolkommission (1996). Note that we cannotassign both competition measures perfectly to each firm for tworeasons. Firstly, for the construction of this measure, the AntitrustCommission uses information on firms’ sales in individual marketsegments. Hence, there are several competition measures for eachfirm depending on sales structure. Unfortunately, our main sourceof data, the Hoppenstedt database, assigns firms only to oneindustry, the primary product market. Hence, our competitionmeasure may contain some classification error for large firms.


Secondly, the classification of industries used by the AntitrustCommission differs from the industry classification used in theHoppenstedt database (European NACE code). Therefore, we hadto assign some firms on an individual basis.

FIRM-LEVEL COMPETITION

The construction of our measure for firm-specific rents (RENT) isdiscussed in Section 3Ð2.

CORPORATE GOVERNANCE: OWNERSHIP CONCENTRATION

The construction of our preferred measure for ownership concen-tration (CONTROL), as well as two alternative measures (BLOCKand HERF) are discussed in Section 3Ð3.

CORPORATE GOVERNANCE: TYPE OF OWNER

We classify firms into five ownership categories (TYPE): pri-vate (including partnerships and foundations), financial firms(including banks and insurers), non-financial firms, governmentauthorities. If a firm has no ultimate owner according to the def-inition of CONTROL as outlined in Section 3Ð3, the ownershipcategory is ‘‘dispersed’’.

CORPORATE GOVERNANCE: OWNERSHIP COMPLEXITY

Ownership complexity is measured with an indicator variable forcross-ownership (CROSS). It takes the value of one if a firm’sultimate owner is part of the web of industrial and financialGerman firms identified by Wenger and Kaserer (1998) and ifthe ultimate owner indirectly owns a share block in itself, zerootherwise.

Documents

Product market competition, corporate governance and firm performance